Skip to content
BranchesTags

Latest commit

 

History

8 Commits

Folders and files

NameName
Last commit message
Last commit date
.github/workflows
.github/workflows
 
 
scripts
scripts
 
 
.gitignore
.gitignore
 
 
AGENTS.md
AGENTS.md
 
 
CONTRIBUTING.md
CONTRIBUTING.md
 
 
LICENSE
LICENSE
 
 
README.md
README.md
 
 

Repository files navigation

Awesome NVIDIA Licensing

Awesome License: CC0-1.0 PRs Welcome

A curated, practical guide to understanding, deploying, and staying compliant with NVIDIA software and GPU licensing. This repo is designed for both humans (operators, architects, procurement) and coding agents (automated compliance checks, deployment scripts, and policy enforcement).

⚖️ Disclaimer: This is community-curated reference material, not legal advice. NVIDIA license terms change over time and may vary by country, product version, and purchase channel. Always read the current EULA/SLA for the exact product you deploy and consult your legal/procurement teams for binding decisions.

Contents

Quick Reference

Situation License Requirement Enforcement Key Source
Bare-metal CUDA compute on a data-center GPU Generally no license for CUDA driver/runtime itself EULA only CUDA EULA
Access NGC-gated enterprise containers (NIM, Riva, etc.) NVIDIA AI Enterprise required for production NGC entitlement check NVAIE Licensing
vGPU for Compute (time-sliced / MIG-backed) NVAIE license per vGPU instance Software enforced; performance degrades without license vGPU FAQ
RTX Virtual Workstation (vWS) / high-end 3D VDI vWS license per concurrent user (CCU) Software enforced vGPU Client Licensing
Virtual PC (vPC) / multimedia VDI vPC license per concurrent user Software enforced vGPU Client Licensing
Virtual Applications (vApps) / RDSH app streaming vApps license per concurrent user EULA mostly; software for one user per vGPU vGPU Client Licensing
Self-hosted NVIDIA NIM in production NVAIE license per GPU NGC entitlement / EULA NIM docs
GeForce/Titan driver in a data center Prohibited (except blockchain processing) EULA GeForce EULA
NVIDIA Omniverse Enterprise NVIDIA Enterprise entitlement (Omniverse Enterprise included since Oct 27, 2025) License server token (CLS/DLS) Omniverse Licensing
NGC free public containers NGC account / API key for some; no license for many public images Authenticated access for private/entitled content NGC Catalog Guide

Core NVIDIA License Agreements

NVIDIA Software License Agreement (SLA)

Most NVIDIA SDKs, libraries, and enterprise software are governed by the NVIDIA Software License Agreement. It is a click-through agreement you accept when downloading the software.

Key concepts:

  • Licensed, not sold. NVIDIA grants a limited, non-exclusive, non-transferable, non-sublicensable (unless stated) license.
  • Authorized users. Employees, contractors, and (for academic institutions) enrolled/employed users on your secure network.
  • Distribution rights. Only portions explicitly marked as "distributable" may be redistributed, typically as incorporated object code into an application with material additional functionality.
  • Prohibited uses. Reverse engineering, decompilation, stand-alone redistribution, and use beyond the licensed scope.
  • Updates. Maintenance and updates are governed by the same agreement unless superseded.

Agent note: Always locate the exact SLA version embedded with the downloaded artifact (/opt/nvidia/.../LicenseAgreement.pdf or EULA.html).

CUDA Toolkit End User License Agreement (EULA)

The CUDA Toolkit EULA covers the CUDA Toolkit, CUDA Samples, the NVIDIA Display Driver, and NVIDIA Nsight tools.

  • Free to use for development and deployment on NVIDIA GPUs.

  • No per-GPU license fee for the CUDA runtime/driver itself.

  • Export controls apply; confirm you are not in an embargoed jurisdiction.

  • GeForce data-center prohibition is enforced through the driver EULA, not the CUDA EULA per se.

  • CUDA EULA Official

cuDNN, TensorRT, and Deep Learning SDKs

Deep-learning SDKs (cuDNN, TensorRT, TensorRT-LLM, etc.) each have product-specific license supplements.

  • cuDNN is governed by the NVIDIA cuDNN Software License Agreement.

  • TensorRT has its own SLA; newer TensorRT for RTX has a separate EULA.

  • Redistribution is permitted only for portions identified as distributable and only when embedded into an application with material additional functionality.

  • Include the notice: This software contains source code provided by NVIDIA Corporation. when distributing sample-derivative source code.

  • cuDNN SLA Official

  • TensorRT SLA Official

  • TensorRT for RTX License Official

NVIDIA AI Enterprise (NVAIE)

NVIDIA AI Enterprise is the umbrella production license and support subscription for NVIDIA's AI software stack.

What it unlocks

  • Production rights for NIM microservices, Riva, NeMo, Triton Inference Server, TensorRT, cuDNN, and many NGC enterprise containers.
  • Enterprise support (Business Standard or Business Critical).
  • Maintenance, security patches, and long-term branches for selected components.
  • Access to gated NGC content and Helm charts.

Licensing model

  • Per GPU. One license is required for every GPU installed in the server or workstation running NVAIE software.
  • Per server/instance for CPU-only compute environments.
  • Subscription or perpetual. Subscriptions are 1, 3, or 5 years. Perpetual licenses include 5 years of Business Standard Support.
  • Cloud consumption via cloud marketplaces (pay per GPU-hour).

Bundled NVAIE

  • H100 PCIe / NVL and H200 NVL include a 5-year NVAIE subscription (activation required with GPU serial number).
  • A800 40GB Active includes a 3-year NVAIE subscription.
  • DGX Hopper systems include NVAIE in the DGX software bundle.
  • DGX Blackwell (B200/B300) and Blackwell DGX systems do not bundle NVAIE; licenses must be purchased separately and activated via NVIDIA Mission Control / Base Command Manager.

Consolidation note: Effective October 27, 2025, NVIDIA consolidated NVIDIA AI Enterprise, Omniverse Enterprise, and Run:ai self-hosted into a single NVIDIA Enterprise software suite at the same price. Existing NVAIE customers retain access to the consolidated capabilities.

NVIDIA vGPU / Virtual Workstation

NVIDIA vGPU software enables multiple VMs to share a physical GPU. Licensing depends on the vGPU type and use case.

License editions

Edition Target Users Typical Workload
NVIDIA Virtual Applications (vApps) App streaming / RDSH users PC-level apps delivered via Citrix, VMware Horizon, RDSH
NVIDIA Virtual PC (vPC) VDI knowledge workers Full Windows VDI with multimedia and browser video
NVIDIA RTX Virtual Workstation (vWS) Power users High-end CAD, CAE, 3D rendering, professional graphics, AI workloads

Licensing model

  • Per Concurrent User (CCU). One license per active user/session.
  • Annual subscription or perpetual + SUMS (Support, Upgrade, and Maintenance Subscription).
  • vPC and vWS include a vApps entitlement.

vGPU 19.0 and 20.x updates

  • Blackwell support: vGPU 19.0/20.x adds support for RTX PRO 6000 and RTX PRO 4500 Blackwell GPUs, including MIG-backed vGPUs.
  • Density: Up to 48 VMs per GPU on supported configurations.
  • New profile: A 3B vPC profile is available for enhanced virtual PC workloads.
  • End-of-life notice: Starting with NVAIE Infra 8.0 / vGPU 20.x, Tesla V100 and several Quadro RTX / RTX A-class workstation GPUs are no longer supported for vGPU for Compute.

Related toolkits

Enforcement

Deployment Required License Enforcement
A-series vGPU vApps Software (one user per vGPU) + EULA
B-series vGPU vPC or vWS Software
Q-series vGPU vWS Software
GPU passthrough — workstation/3D vWS Software
GPU passthrough — PC apps vApps EULA only
Bare metal — workstation/3D vWS Software
Bare metal — PC apps vApps EULA only

NVIDIA NIM

NVIDIA NIM provides optimized, containerized inference microservices.

Three usage modes

  1. Hosted API catalog (build.nvidia.com)
    • Free tier for prototyping via the NVIDIA Developer Program.
    • Rate-limited; not intended for production traffic.
  2. Downloadable NIM containers
    • Free for development, testing, and research on up to 16 GPUs (Developer Program).
    • Requires NVAIE for production use.
  3. NVAIE production deployment
    • Self-hosted on your own GPUs with full enterprise support.
    • Priced per GPU per year.

Compliance rule of thumb: If real end users or business transactions depend on the output, you are in production and need NVAIE.

Workstation / RTX exception: Some NIM microservices for RTX / GeForce RTX or Grace-Blackwell client systems may be used on a single-user PC or workstation without an NVAIE subscription, subject to the product-specific terms.

NVIDIA NGC and Container Licensing

The NVIDIA NGC Catalog hosts containers, models, Helm charts, and SDK resources.

Access levels

Level View Public Download Public View Entitled Download Entitled Notes
Guest Yes Yes Yes No No account required for many public assets
Authenticated Yes Yes Yes No Requires NGC account + API key
Subscriber Yes Yes Yes Yes Requires NVAIE or product entitlement

Container EULA acceptance

Pulling an NGC container typically constitutes acceptance of the product-specific EULA. Many containers include the EULA at a known path inside the image, e.g. /opt/nvidia/<product>/LicenseAgreement.pdf.

OSS components

NGC containers bundle open-source software. License notices and source-code offers (e.g. for GPL/LGPL components) are usually documented in the container or on the NGC page.

NVIDIA License System (NLS)

NLS is the modern entitlement-delivery platform for NVAIE, vGPU, and Omniverse Enterprise.

Service instance types

Type Hosting Use case Internet required?
Cloud License Service (CLS) NVIDIA-managed in the cloud Internet-connected clients; easiest setup Yes (egress 80/443)
Delegated License Service (DLS) On-premises virtual appliance Air-gapped / sovereign / restricted networks No, for clients

High-level workflow

  1. Purchase entitlements through NVIDIA or a reseller; receive a PAK ID (30-character entitlement key).
  2. In the NVIDIA Licensing Portal (nvid.nvidia.com), create a License Server.
  3. Create or register a Service Instance (CLS or DLS).
  4. Bind the License Server to the Service Instance.
  5. Install the license on the service instance.
  6. Generate a Client Configuration Token and deploy it to licensed clients.
  7. Clients lease licenses from the service instance at boot and return them at shutdown.

Legacy license server

Older vGPU releases used the NVIDIA vGPU Software License Server (FlexNet-based). New deployments should use NLS (CLS/DLS).

Cloud and Marketplace Licensing

Public cloud options

Model License handling When to use
Pay-as-you-go (PAYG) License included in hourly per-GPU price Burst, experimentation, no existing NVAIE
Bring Your Own License (BYOL) Use existing NVAIE subscription on certified cloud instances Already own NVAIE, need consistent entitlements
Cloud marketplace subscription Billed through CSP marketplace Align licensing with cloud procurement

Important notes

  • H100/H200 NVL cloud instances may include NVAIE depending on the provider offering.
  • B200/B300 cloud instances typically require separate NVAIE purchase.
  • GPU passthrough VMs in the cloud follow the same rules as on-prem: NVAIE is required only for NVAIE-gated software, not for basic CUDA.
  • vGPU for Compute in the cloud is software-enforced and requires NVAIE per vGPU instance.

Hardware-Specific Restrictions

GeForce and Titan in data centers

NVIDIA's GeForce/Titan driver EULA includes the clause:

No Datacenter Deployment. The SOFTWARE is not licensed for datacenter deployment, except that blockchain processing in a datacenter is permitted.

Practical implications:

  • Consumer-grade GPUs (GeForce/Titan) are not authorized for data-center software deployments.
  • This restriction applies to the software/driver, not ownership of the hardware.
  • Non-commercial, small-scale research/LAN use that does not operate at data-center scale is generally not the target of enforcement, per NVIDIA public statements.
  • Datacenter/colocation providers often reject consumer GPUs for this reason.

H100 / H200 / A800 bundled NVAIE

Selected GPUs include NVAIE subscriptions activated by serial number. Track serial numbers and activation dates for compliance audits.

DGX systems

  • DGX Hopper includes NVAIE in the DGX software bundle.
  • DGX Blackwell (B200/B300) and GB200/GB300 NVL72 require separately purchased NVAIE / NVIDIA Enterprise licenses, activated via NVIDIA Mission Control / Base Command Manager.

Open Source Software (OSS) Compliance

NVIDIA products frequently incorporate open-source components (Linux kernel modules, drivers, libraries, etc.).

What to track

  • Component inventory: Name, version, origin URL, license name.
  • Linking/dynamic vs. static: Static linking of copyleft libraries can affect your obligations.
  • Modifications: Any changes to OSS components must be recorded.
  • Distribution terms: Permissive licenses (MIT, Apache-2.0, BSD) usually require attribution; copyleft licenses (GPL, LGPL, AGPL) may require source publication.

NVIDIA open-source resources

  • NVIDIA GPU kernel modules are open-source (nvidia/open-gpu-kernel-modules).
  • CUDA and driver packages include OSS notices and source-code offers.
  • Container SBOMs and VEX documents are available from NGC for vulnerability and license review.

Practical compliance steps

  1. Generate or retrieve SBOMs for every NVIDIA container you ship.
  2. Review licenses of bundled components before redistribution.
  3. Preserve copyright notices and license texts.
  4. For GPL/LGPL components, honor source-code offer requirements.
  5. Use license-scanning tools (FOSSology, ScanCode, ORAS) in CI/CD.

Kubernetes and GPU Operator Licensing

The NVIDIA GPU Operator automates driver, container runtime, and device-plugin deployment on Kubernetes. Licensing considerations:

Run:ai note: Starting with NVAIE Infra 7.5 and continuing in 8.x, Run:ai SaaS is included alongside the self-hosted option under the NVIDIA Enterprise suite.

Node type NVAIE required? Notes
Bare-metal worker nodes Only for NVAIE-gated NGC software Basic CUDA workloads do not need NVAIE
vGPU-backed VM worker nodes Yes, per vGPU instance GPU Operator can distribute NLS client tokens automatically
GPU passthrough VM worker nodes Same as bare metal inside the VM
CPU-only nodes running NVAIE software One NVAIE subscription/license per server/instance

NLS client configuration token paths

Licenses are obtained via a Client Configuration Token (.tok file) that identifies the client to CLS/DLS.

OS Default token directory
Linux /etc/nvidia/ClientConfigToken/
Windows %SystemDrive%:\Program Files\NVIDIA Corporation\vGPU Licensing\ClientConfigToken\

Ensure:

  • The directory contains only the token intended for the client.
  • File permissions are 744 (Linux) or equivalent restricted access.
  • The nvidia-gridd service (Linux) or NVIDIA Display Container LS service (Windows) is restarted after token deployment.

Common Licensing Mistakes

  1. Assuming CUDA requires a license. The CUDA toolkit/driver does not require a separate license fee; compliance is EULA-based.
  2. Using free NIM endpoints for production. The hosted build.nvidia.com endpoints are rate-limited and intended for prototyping.
  3. Forgetting vGPU licensing scales by concurrent user. A vPC/vWS/vApps license is needed per active session, not per physical GPU.
  4. Deploying GeForce/Titan in data centers. The driver EULA explicitly prohibits data-center deployment except for blockchain processing.
  5. Not tracking bundled NVAIE activation. H100/H200/A800 GPUs include NVAIE subscriptions tied to serial numbers and ship dates.
  6. Mixing CLS and DLS in HA. High-availability clusters must be composed entirely of DLS instances; CLS and DLS cannot be mixed in the same HA cluster.
  7. Ignoring OSS obligations in NGC containers. Containers include GPL/LGPL and other components; retain SBOMs and honor source-code offers.

Compliance Checklists

Pre-deployment

  • Identify every NVIDIA software component in the stack (driver, CUDA, cuDNN, TensorRT, NIM, vGPU, GPU Operator, etc.).
  • Download and read the exact EULA/SLA version bundled with each component.
  • Confirm GPU model and deployment mode (bare metal, passthrough, vGPU, cloud PAYG/BYOL).
  • Determine whether NVAIE is required and how many GPU licenses are needed.
  • Verify GeForce/Titan hardware is not deployed in a data-center driver scenario.
  • Choose CLS vs. DLS based on network topology and security requirements.

Operational

  • Register all GPU serial numbers that include bundled NVAIE.
  • Keep license-server uptime and HA monitoring in place (especially DLS).
  • Monitor license usage vs. purchased entitlements.
  • Renew subscriptions before expiration to avoid service degradation.
  • Document user access and authorized-user policies.

Audit / review

  • Maintain an inventory of NVIDIA entitlements, PAK IDs, and license servers.
  • Retain SBOMs and OSS notices for redistributed containers.
  • Review rate-limit and production-use clauses for hosted NIM endpoints.
  • Validate that vGPU workloads match purchased CCU counts.
  • Run ./scripts/validate-links.sh before publishing doc updates.

Tools and Official Resources

NVIDIA portals and docs

Compliance / SBOM tools

  • ORAS CLI Community — fetch NGC SBOM and VEX artifacts.
  • FOSSology Community — open-source license and copyright scanner.
  • ScanCode Toolkit Community — license and origin scanner.
  • SPDX Community — standard format for license data exchange.

License-server utilities

  • nvidialsadmin Official — CLI for managing DLS/CLS service instances.
  • NVIDIA License System User Guide Official — referenced above.

Base Command Manager and Mission Control

Glossary

Term Meaning
CCU Concurrent User — a vGPU license metric.
CLS Cloud License Service — NVIDIA-hosted license delivery.
DLS Delegated License Service — on-premises license delivery.
EULA End User License Agreement.
NVAIE NVIDIA AI Enterprise.
NLS NVIDIA License System.
NGC NVIDIA GPU Cloud / NVIDIA NGC Catalog.
NIM NVIDIA Inference Microservices.
PAK ID Product Activation Key — 30-character entitlement identifier.
SBOM Software Bill of Materials.
SLA Software License Agreement.
SUMS Support, Upgrade, and Maintenance Subscription.
VEX Vulnerability Exploitability eXchange document.
vGPU NVIDIA Virtual GPU technology.
vPC / vWS / vApps Virtual PC, RTX Virtual Workstation, Virtual Applications.

Contributing

Read CONTRIBUTING.md for the quality bar, entry format, and PR process.

License

This list is released into the public domain under CC0-1.0.

Want us to build this for you?

Enterprise AI Atlas is maintained by Vibe Coding Agency. We prototype and ship agentic systems, MCP servers, and enterprise AI integrations for teams that need working software fast — without hiring a full AI engineering team.

Free guide: The Non-Technical Founder's Guide to Agentic AI — what agents and MCP servers are, and how to get a system built.

About

Curated NVIDIA software licensing resources, compliance guides, and product entitlements.

Topics

awesome-listcomplianceenterprise-aigpu-softwarenvidianvidia-licensingsoftware-licensing

Resources

Readme
CC0-1.0 license

Contributing

Contributing
Activity
Custom properties

Stars

0 stars

Watchers

0 watching

Forks

0 forks
Report repository

Releases

Packages

Contributors

Languages